Experimental Characterization of a

Femtocell Radio Channel

João C Braz1, Pedro Gonzalez Castellanos

1, Carlos Rodriguez Ron

2, Luiz A R Ramirez

2, Leonardo Gonsioroski

2,

Luiz da Silva Mello1,2

, Flávio Hasselmann2

1Instituto Nacional de Metrologia, Qualidade e Tecnologia, INMETRO, Xerém, Brazil

jabraz@inmetro.gov.br 2Pontifical Catholic University of Rio de Janeiro, PUC-Rio, Rio de Janeiro, Brazil

smello@cetuc.puc-rio.br

Abstract— Propagation measurements performed at 1.95 GHz

with a 160MHz channel bandwidth in an indoor environment are

presented and compared with theoretical predictions obtained

using the Saleh-Valenzuela model and FDTD simulation.

Keywords-channel modeling; channel sounding; FDTD; femtocells

I. INTRODUCTION

This paper reports measurements and channel modeling for a conference room located in a building at the Catholic University of Rio de Janeiro (PUC-Rio) that is an example of a femtocell, a special indoor environment with limited penetration of signals from outdoor radio base stations. Appropriate location of femtocell devices, in order to guarantee the best quality of service, requires knowledge of Power Delay Profiles (PDP) obtained in a sound radio channel characterization for the wideband channels.

Propagation in radio channels with spatial and temporal selectivity can be characterized using the Saleh and Valenzuela model [1], based on the assumption that multiple signals reaching the receiver arrive in clusters and each cluster consists on many rays. In this paper, the measured PDP results are used to extract the Saleh-Valenzuela model parameters for the environment and compared with results obtained with FDTD [2] simulations.

II. MEASUREMENT SETUP

The environment for the indoor measurement campaign is a 12.3 x 15.0 x 8.0 m conference room, illustrated in Fig. 1. A set of eight fixed reception points (RX1 to RX8) were selected near different types of materials like wood, glass and metal, on LOS (line of sight) condition at different distances from the transmitter. The sounding technique used time domain measurements of PN sequences with implementation of the matched filter by software at the receiver end [3]. A sounding bandwidth of 160 MHz allowed a spatial resolution of 3.75 m, with maximum delay of 3.1875 µs and a theoretical dynamic range of 48 dB.

Both the transmitter (TX) and receiver antennas employed are omnidirectional discones with an approximately constant return loss about 15 dB over the frequency range of interest, operating in vertical polarization.

The generated driving signal in (1) has a central frequency (fc) of 1.95 GHz over a band (fb) of 80 MHz and t0 = 12.5 ns denotes the initial delay; signal (normalized) amplitude and its frequency contents are illustrated in Fig. 2.

E

z(t) = E

0 sin [2π f

c (t - t

0)] exp[-(2 f

b (t - t

0))2] (1)

Figure 1. Conference room on PUC-Rio Building.

Figure 2. Pulse excitation of TX antenna and its frequency spectrum.

III. CHANNEL MODELLING

A. Small scale variations In order to analyze a statistic behavior of small-scale

amplitude variations, the received signal amplitudes CDF was compared with the Rician and Nakagami distributions. Fig. 3 shows the results, indicating that the Nakagami distribution

This work is supported by CNPq, under covenant 573939/2008-0 (INCT-

CSF), and by FAPERJ under fellowships E-26/100.078/2008 and E-

26/102.407/2010.

6th European Conference on Antennas and Propagation (EUCAP)

978-1-4577-0919-7/12/$26.00 ©2011 IEEE 2064

provides a much better fit. The adjusted distributions parameters are shown in Table I.

Figure 3. Fitted distributions.

TABLE I. Distribution Fitting Parameters

Parameter Nakagami Rician

m 0.21

Ω 1.9

s 0.00012

σ 0.74

B. Saleh and Valenzuela model The channel impulse response for the wideband channel can

be represented by the Saleh-Valenzuela model [1]:

h(t) = βkl

ejϕklδ(t − T

l− τ

kl)

k=0

∞

∑l=0

∞

∑ (2)

where βkl is the amplitude of the channel response, φkl is the phase component of the k-th ray inside of l-th cluster. The complex part e

jφkl represents a statistically independent random

phase associated with each arrival. It is assumed that φkl is uniformly distributed in the interval [0, 2π]. Tl represents the time of arrival of the l-th cluster and the time τkl to get the k-th ray within the l-th cluster. A graphical representation of the model is shown in Fig. 4.

Figure 4. Graphical representation of the Saleh and Valenzuela model.

The arrival times are assumed to be exponentially distributed

( )[ ]1)|( 1−−Λ−

− Λ= ll TTll eTTp (3)

( )( )( )[ ]lkklep lkl

1)|( 1−−−

− = ττλλττ (4)

The amplitudes are defined as statistically independent

random variables with mean square value βkl2 being a

monotonically decreasing function of Tl and τl:

β

kl2 = β2(0;0)e

−Tl /Γe−τl /γ (5)

Equation (5) represents the double decay of Fig. 1, with Γ and γ defining the constant decay of cluster and rays,

respectively, and β2(0;0) representing the amplitude of the first

ray of the first cluster. A set of coefficients of the Saleh and Valenzuela model

was obtained for each measured profile. Then, the parameters of the statistical distributions of each coefficient were obtained. To characterize this type of environment the required parameters are the mean number of clusters, the inter-cluster

arrival rate Λ, the inter-ray arrival rate λ, the inter-cluster delay constant Γ and the cluster decay time constant γ.

Tabel II summarizes the values obtained. Also shown are the values recommended in the IEEE802.15a standard [4] for indoor residential, indoor office and industrial environments.

TABLE II. MODEL PARAMETERS

Parameter Conference

room

Indoor

residential

Indoor

office

Industrial

<L> 2 3 5.4 4.75

Λ (1/ns) 0.0163 0.047 0.016 0.0709

λ (1/ns) 0.052 1.54 0.19 NA

Γ (ns) 70 22.61 14.6 13.47

γ (ns) 19 12.53 6.4 0.651

C. FDTD implementation The FDTD method was also tested for simulating the

measured PDP´s. For the (12.32 x 15 x 8) m scenario at hand and a spatial

step of 0.0154 m, a total of 557573118 FDTD cells are represented, each cell associated to six (electric and magnetic) field values as well as to four flags indicative of the type of material present at each spot. Also, field values are 4 bytes real numbers while flag lengths are 1 byte each, implying a memory need of 30 bytes per cell and a total memory requirement of about 16.72 GB to address the present problem, which was perfectly handled by the supercomputer of CESUP/UFRGS of 324 GB (RAM) processors where the application was running.

The CUDA FDTD version of the code was changed to work on GPU Tesla S1070 with 240 kernels, 4 Teraflops, clock rate 1.44 GHz and 16 GB of global memory. The (x,y,z) E and H field components were stored on the device memory as 32 bit floating point variables. A texture memory was used to store, as a pointer stream, the material types in the model space.

2065

In both of these cases, the 3D volume was flattened into 2D and was accessed via an algorithm based on 3D to 2D address translation. This allows the entire 3D space to be updated in one render pass and also avoids potential "read after write" data corruption. The material type pointer stream was used for material property lookups stored in textures. E and H scattered field update calculations were converted to fragment programs, taking the E and H fields values stored in textures as inputs.

The constitutive parameters of different materials of the

environment are input parameters that are taken into account

in the simulation algorithm. The values in Table III are

associated to different colours in Fig. 5. The transmitter and

receivers antennas positions are also indicated in Fig. 5, by TX

and RXn.

TABLE III. CONSTITUTIVE PARAMETERS OF THE MATERIALS.

Material Relative

permittivity

Conductivity

(S/m)

Wood 3,0 0,001

Concrete 6,0 0,05

Glass 2,7 0,008

Metal 1,0 1,0x109

Figure 5. Points of measurement (different materials).

Measured and FDTD-simulated PDP´s are shown in Figs. 6 and 7, respectively.

Figure 6. Measured power delay profiles.

Figure 7. FDTD simulated power delay profiles.

IV. SIMULATION RESULTS

In order to analyze the suitability of the Saleh – Valenzuela model to describe the channel, 1.000 power delay profiles were generated and the convergence of this simulation was tested.

Fig. 8 shows the RMS delay spread CDF’s obtained from the unfiltered measured data and from the data filtered using the Constant False Alarm Rate (CFAR) technique [5] that eliminates false multipath components due to noise. Also shown are the CDF’s obtained by simulation of 100, 500 and 1.000 profiles. It leads to the conclusion that 500 PDP´s were sufficient to generate the channel response.

The average RMS delay spread obtained with the Saleh-Valenzuela model was 14 ns and the corresponding value obtained from the measurements was 15 ns.

2066

Figure 8. Convergence evaluation of Saleh-Valenzuela model.

In Fig. 9, the measured results are the compared with the FDTD simulation. It can be observed that there is a large discrepancy between the CDF’s of RMS delay spread. One of the reasons for this discrepancy may be that the real environment was not sufficiently detailed on the simulation process. The FDTD simulation estimates of the RMS delay spread shows the presence of more scatters than were actually measured, indicating the need for a best description of the environment in terms of electrical characteristics of the materials and objects in the environment. On the other hand, this may also be due to the limited resolution of the experimental setup (3.45 m).

Figure 9. Simulated and measured CDF´s of RMS delay.

V. CONCLUSIONS

An efficient sounding technique was implemented to test the behavior and help to characterize the radio propagation of a single indoor channel.

The parameters of the Saleh-Valenzuela model for the channel were obtained. The model was validated by the simulating 1000 profiles with the extracted parameters and comparing the CDFs of RMS delay spread for measured and simulated profiles. A good agreement was observed. The average RMS delay spread obtained with the Saleh-Valenzuela model was 14 ns and the corresponding value obtained from the measurement was 15 ns.

FDTD simulation showed significant differences between the values of RMS delay spread and those obtained in the measurements. This is attributed to the lack of detail in the environment characterization.

The results are useful to the design of femtocell systems to operate in indoor environments. Work in progress comprehends new measurement campaigns in different scenarios in order to improve the simulation results.

ACKNOWLEDGMENT

The authors are indebted to CESUP/UFRGS for allowing us to use the supercomputer and Tesla GPUs.

REFERENCES

[1] A. Saleh, and R. Valenzuela, “A statistical model for indoor multipath

propagation,” IEEE Journal on selected areas in communications. vol.

SAC-5, pp. 128-137, 1987.

[2] J. A. Taflove, and S. C. Hagness, Computational Electrodynamic: the Finite-Difference Time-Domain Method, 3rd Ed., Arthec House

Publishers, 2005.

[3] J. A. Cal Braz, “The Relevance Vector Machine Applied to the

Modeling of Wireless Channels”, M.Sc Dissertation, UFF, Rio de Janeiro, 2010 (in Portuguese).

[4] A. F. Molisch et al., “IEEE 802.15.4a Channel Model – Final Report,”

Tech. Rep., Document IEEE 802.15.04-0662-02-004a, 2005.

[5] E. Souza, V. Jovanovic, and C. Daigneault, “Delay spread measurements for the digital cellular channel in Toronto," IEEE Transactions on

veicular tecnhnology, vol 43. No 4, pp. 837-847, 1994.

2067